New Annual Review with @nathanieldaw.bsky.social: “Planning in the Brain: It's Not What You Think It Is.” We argue that the brain's 'planning' machinery is mostly used for learning from simulated experience, and that thinking prospectively at decision time is just one special case of this process.
New work w/ Zach Kelso and @madeleinecsnyder.bsky.social
www.biorxiv.org/content/10.6...
Our negative results on classical conditioning in planarian flatworms. This was surprising, given the long history of work (including sensational findings of memory transfer and retention through decapitation).
We're happy to release NeuralSet: a simple, fast, scalable package for Neuro-AI
Supports:
🧠 fMRI, EEG, MEG, iEEG, spikes… preprocessing
💬 text 🔊 audio ▶️ video 🏞️ image… embeddings
📦 pip install neuralset
🔍 facebookresearch.github.io/neuroai/neur...
📄 kingjr.github.io/files/neural...
🧵 Details👇
Repeated lifestyle changes drove the unique evolution of vertebrate eyes. Cross-section diagrams of likely photoreceptor (PRC) and eye structures in the heads of ancestral bilaterians (top), with presumed ancient lifestyles (bottom). Approximate times in million years before present are indicated for key evolutionary stages.
How eyes on modern vertebrae came to be is amazing. There are really complex; it's a multilayered circuit of rods, cones, and rhabdomeric photoreceptors. Getting there was just as complex an evolutionary journey. 🧪
Link: www.cell.com/current-biol...
To accompany my textbook (Computational Foundations of Cognitive Neuroscience) and the class I taught this semester, I'm open-sourcing my lectures slides:
gershmanlab.com/lectures.html
I'll continue to update these as I improve them.
I think this is a super interesting paper from @markhisted.org and co:
www.cell.com/neuron/abstr...
My personal bet is that the phenomenon seen here would be different for some types of inter-laminar recurrence.
#neuroscience 🧪
Going from neural activity to blood flow just became easier! Two brainwide populations, each with its neurovascular coupling. (But going backwards... is now a tad more complicated.)
By @agnesland.bsky.social & team.
Thanks @intlbrainlab.bsky.social @wellcometrust.bsky.social @simonsfoundation.org
We didn't look at this in detail, but yes CS dopamine response was also correlated with the number of consumption licks. The US dopamine response was more predictive of this though (which makes sense since the CS was at that point 3s ago)
We used TD models to show that these correlations are best explained by a model where cue-evoked dopamine (putatively the cue-evoked RPE) has a direct role in modulating responding. This suggests that dopamine, in addition to its role in driving learning, also directly modulates responding. (🧵8/8)
These correlations were present across multiple previously published studies, and we also show this is consistent with the results of previous optogenetic perturbation studies. (🧵7/8)
We found strong trial-to-trial variability between the dopamine response to the cue (CS) and the subsequent anticipatory lick rate in mice during a standard trace conditioning paradigm, both throughout and at the end of learning. (🧵6/8)
A typical assumption is that anticipatory responding reflects the animal's estimate of value. But recent work suggests dopamine/RPEs might play a role as well! Here we set out to disentangle the relationship between value, dopamine/RPEs, and anticipatory licking. (🧵5/8)
One thing that TD learning does NOT specify, however, is animals' anticipatory/conditioned responding—e.g., mice lick the water spout upon delivery of a reward-predicting odor. Mice show trial-to-trial variability in their anticipatory response rates that is not well understood. (🧵4/8)
Learning is thought to involve updating value estimates using a reward prediction error (RPE). The phasic activity of dopamine neurons looks a lot like the RPE signal, suggesting dopamine is involved in learning value. (🧵3/8)
In Pavlovian conditioning, animals learn to respond to a conditioned cue that predicts future reward. This learning process can be understood using temporal difference (TD) learning, which assumes that animals estimate the "value" of the conditioned cue. (🧵2/8)
New paper in collaboration with @mhburrell.bsky.social @naoshigeuchida.bsky.social and @gershbrain.bsky.social !
"Phasic dopamine drives conditioned responding beyond its role in learning" 🧵 (1/8)
www.biorxiv.org/content/10.6...
Someone needs to make a calculator where the eml operator is the only operator button. This is genuinely very cool though
The basal ganglia famously receive RPE-like input from dopamine neurons. However, dopamine neurons project much more to the striatum than the cortex. Does the cortex receive a complementary learning signal from a different neuromodulator? We think NE is one clear candidate. 13
Delighted to share our discoveries about one of the brain's neurotransmitter systems:
www.biorxiv.org/content/10.6...
Together with colleagues at the @alleninstitute.org, we have learned a lot about a tiny cluster of neurons in the brainstem locus coeruleus (LC) that releases norepinephrine (NE). 1
New preprint! 🧠
How do RNNs learn abstract rules from sequences, independent of specific stimuli?
By Vezha Boboeva, with Alberto Pezzotta & George Dimitriadis
"From sequences to schemas: low-rank recurrent dynamics underlie abstract relational representations"
www.biorxiv.org/content/10.6...
@jhennig.bsky.social has shown that dopamine exerts a real-time effect on conditioned responding, beyond its role in learning:
www.biorxiv.org/content/10.6...
Another indication that dopamine is more than a learning signal!
A joint effort with @naoshigeuchida.bsky.social and @mhburrell.bsky.social.
If you're interested in emerging ideas in neural interfaces, I humbly suggest my lab's latest: www.nature.com/articles/s42...
Neural interfaces create dynamic interactions between the brain & devices. This means mean we need new engineering approaches beyond typical ML to "decode" a static brain
How similar are two activity patterns? We all know that low correlations don't mean that they are necessarily dissimilar – our measurements could be noisy. So how do you correct for this? Our slightly nerdy, but hopefully useful preprint takes a deep dive:
www.biorxiv.org/content/10.6...
Me too! And you just knew they had to have a 7/4 song somewhere!
Also whenever we're ready for 9/8...
Wayne Shorter - Ponta de Areia www.youtube.com/watch?v=VFPI...
Ooh possum kingdom is a good one! I found two more on a road trip but sadly forgot one of them 😭
Stereolab - Motoroller Scalatron www.youtube.com/watch?v=vlN4...
Thrilled to share our new paper, which shows that the relative timing of cholinergic and dopamine release dynamically gates whether dopamine acts as an RPE for in vivo plasticity and reinforcement learning. www.nature.com/articles/s41...
Have you ever wondered why mice do what they do when they are free to do whatever they want? Check out our latest (and this slightly delayed thread about our recent paper, led by Caleb Weinreb and friends...) www.cell.com/neuron/fullt...
If there were one fact I wish more people in the field knew, it's that randomly-initialized RNNs can do basically anything (as in reservoir computing)
🧵 New preprint led by @bingbrunton.bsky.social, @elliottabe.bsky.social, @lawrencehu.bsky.social
We gave a worm brain control of a fly body and it walked
What did we learn? Nothing, other than deep reinforcement learning is effective
We call it the digital sphinx
www.biorxiv.org/content/10.6...
